What environmental, social or economic factors identify high-value wetlands? Data-mining a wetlands database from south-eastern Australia

Abstract

Large amounts of potentially useful information are collected by management agencies as they attempt to identify high-value wetlands and rank them for investment, protection or rehabilitation. Resource constraints frequently mean these information-rich databases are not fully interrogated, with the result that much of their expensively obtained information is only partially analysed or, worse, is not analysed at all. The present paper shows the benefit of rigorously interrogating such databases to identify wetlands of high social, economic or environmental value. Three data-mining methods, namely, univariate analysis, multivariate analysis and artificial neural networks (ANNs), were applied to a large (7.6 MB) but hitherto unanalysed database of 163 wetlands in the Gippsland region of south-eastern Australia. Simple statistical techniques, such as univariate analysis and binary logistic regression, identified high-value wetlands with a prediction accuracy of >90%, using only a small set of environmental indicators. Artificial neural network models with nine environmental-value inputs (six direct indicators plus three threat indicators) correctly also identified 90% of high-value wetlands. Outcomes generated by ANNs were in close agreement with those obtained with more traditional univariate and multivariate analyses. There seems little justification for undertaking economic assessments, and for environmental assessments the best indicators consistently included the presence of listed fauna or flora, vegetation intactness and the absence of hydrological modification. The overall approach, although developed from the analysis of a single (but large) wetland database of wetlands in south-eastern Australia, is likely to find conservation applications in many other regions of the Pacific.